Negative Human Rights as a Basis for Long-term AI Safety and Regulation

Negative rights offer a practical framework for AI safety: The paper proposes that negative human rights—which prohibit harmful actions rather than mandate positive ones—can serve as clear, implementable principles for constraining AI behavior in unpredictable situations, bridging the gap between abstract safety goals and technical requirements.

Legal precedent provides a tested model: By drawing from established human rights law, the approach leverages centuries of legal development and international consensus, making it more likely that AI safety principles based on negative rights would gain widespread acceptance across different countries and cultures.

Specificity enables technical implementation: Unlike vague safety concepts, negative rights are concrete enough to be translated into actual code and algorithms, allowing AI developers to build these constraints directly into systems rather than relying solely on external oversight.

International regulation requires shared foundational principles: The paper argues that binding global AI regulation will only work if countries agree on underlying principles first—and negative human rights, already embedded in international law, could provide that common ground for coordinated governance.

This approach addresses both current and future AI risks: The framework is designed to help AI systems recognize and avoid harmful behaviors not just in scenarios developers anticipated, but in genuinely novel situations where the system must apply general principles to make safe decisions.